In the realm of personalized music recommendations, understanding the mood conveyed by an audio file can significantly enhance the user experience. This project proposes a novel approach to song recommendation by leveraging voice tone and speech-to-emotion detection. Using a dataset of Spotify music data combined with voice recordings, we will build a model to analyze and classify the mood of audio files based on their extracted features.

The dataset provided by Kaggle includes various musical features such as 'danceability', 'acousticness', 'energy', 'instrumentalness', 'liveness', 'valence', 'loudness', 'speechiness', and 'tempo'. Additionally, we will incorporate speech-to-emotion analysis to detect the emotional tone from voice recordings. This involves analyzing audio features to determine emotions such as happy, sad, energetic, and calm.

To achieve this, we will employ advanced machine learning models, including Random Forest, XGBoost, and Gradient Boosting, for mood classification based on the musical features. For speech-to-emotion detection, we will use Convolutional Neural Networks (CNN) and Long Short-Term Memory networks (LSTM) to capture and interpret the emotional content of voice recordings.

Our system will first analyze the given voice or audio file, extracting relevant features from both the music and speech components. The mood classification model will then determine the emotional tone, and based on this analysis, the system will recommend songs that align with the user's current emotional state. This approach aims to deliver a more intuitive and emotionally resonant music recommendation experience by integrating both music and speech emotional analysis.

Keywords: Voice tone analysis, speech-to-emotion detection, mood classification, music recommendation, Spotify data, machine learning, Random Forest, XGBoost, Gradient Boosting, CNN, LSTM, audio features, emotional state.